Commit Graph

8509 Commits

Author SHA1 Message Date
Gregory Hainaut f76bf9dddc gsdx: dump and log EE texture read
It gives a visual opportunity to detect a bad read of the texture cache
2016-07-14 19:41:21 +02:00
Gregory Hainaut 10ea05bc6f common: remove old atomic wrapper
Use cross-platform std::atomic instead
2016-07-14 18:29:41 +02:00
Gregory Hainaut 812e41d578 common: relax atomic of m_IsBeingDeleted
Avoid the memory fence in the constructor
2016-07-14 18:29:41 +02:00
Gregory Hainaut 086dfc8a14 gsdx sw: use acquire/release semantics for atomic operation
* Avoid the generation of memory barrier (mfence)
2016-07-14 18:29:41 +02:00
Gregory Hainaut 567976e822 MTVU: port vuCycles to std::atomic
V2: use relaxed order as the variable doesn't carry load/store dependency
It is only used as a counter for the speed hack
2016-07-14 18:29:41 +02:00
Gregory Hainaut 3f0655c821 MTVU: port write pointer to atomic and optimize atomic access
Write pointer can be relaxed-read from the EE thread
Read pointer can be relaxed-read from the VU thread

Warning: AtomicExchange calls were replaced by release-store
However AtomicExchange generates a memory barrier (at least on linux)
So either it is useless and it will be faster, or it will explode ;)
2016-07-14 18:29:41 +02:00
Gregory Hainaut ca46921796 MTVU: use acquire/release semantics for atomic operation
* Avoid the generation of memory barrier (mfence)
* Based on the fact that it used to work on previous code without any
  barrier

v2:
* keep basic code in reset path
* use relaxed access for isBusy. The variable doesn't carry load/store
  dependency but is instead an hint to optimize semaphore post
2016-07-14 18:29:40 +02:00
Gregory Hainaut 3b4c357aaa MTGS: avoid a potential very rare deadlock 2016-07-14 18:29:40 +02:00
Gregory Hainaut 4cdf05efac MTGS: use private member for RingBufferLock
Easier to understand the purpose of the 2nd lock
2016-07-14 18:29:40 +02:00
Gregory Hainaut 5913ff1bd8 MTGS: port read/write pointer to full atomic + optimize atomic access
Read pointer is only written by the MTGS thread so it can be relaxed-read
by the MTGS thread

Write pointer is only written by the EE thread so it can be relaxed-read
by the EE thread

Remove volatile stuff
2016-07-14 18:29:40 +02:00
Gregory Hainaut 6fb7beb58a MTGS: use acquire/release semantics for atomic operation
* Avoid the generation of memory barrier (mfence)
* Based on the fact that it used to work on previous code without any
  barrier

v2:
* keep "slow" and safe atomic on reset path
* use relaxed atomic for m_RingBufferIsBusy
The signal doesn't carry dependency with others load/store. Instead it is used
as an hint to awake the semaphore
As a side, there is a potential thread bug

T1 do
* wait sema
* busy = true;
* while (!queue.empty) do work...
* busy = false;
* go back to wait sema

T2 do
* post sema even if busy is false

If T1 stop after the while queue but before the busy, T2 won't post the
event. When T1 will wake up, it will block on the semaphore
2016-07-14 18:29:40 +02:00
Gregory Hainaut d855bc5ca8 gsdx sw: improve exit condition of SW extra thread
Use a relaxed atomic to read the exit variable in the hot path

Wait that exit is deasserted in the destructor, so we are sure the
thread will "soon" return
2016-07-14 18:29:22 +02:00
Gregory Hainaut abc9f7d096 gsdx tc: log unsupported LookupDepthSource 2016-07-14 18:29:22 +02:00
ramapcsx2 093704f073 Merge pull request #1456 from NZJenkins/master
Zero out the skip bp
2016-07-12 12:15:09 +02:00
NZJenkins b491f5cdc5 Zero out the skip bp
Fixes #1453
2016-07-12 17:51:56 +12:00
Gregory Hainaut 91eccb7bab gsdx sw: increase the size of the ring buffer 256 => 65536
memory overhead by thead is only 256KB

However it will reduce the probability to block the push thread to nearly 0

I tested a couple of dumps and only manage 4000 element with 1 extrathread.
2016-07-10 10:49:06 +02:00
Gregory Hainaut 85fb55a0e1 gsdx ogl: memory management take 3
Add a factor 2 on the VRAM to get the quantity of available memory for the textures.
The driver is allowed to put some textures in RAM. Of course it is bad for performance
but it won't crash.

Due to the 4GB by process limit, I keep a (reasonable) maximum of 3.8GB.

In order to avoid a crash when memory is too low an exception will be risen
with no guarantee on rendering and big performance impact. In this situation
you ought to reduce upscaling/disable large framebuffer.
2016-07-10 10:23:23 +02:00
Gregory Hainaut a37cd40ce3 gsdx ogl: only print an error when VRAM is low
Until a better solution is found or people buy better GPU :)
2016-07-09 11:43:28 +02:00
Gregory Hainaut a2086ed458 gsdx profiler: drop latest frame time
Value is out of the chart. I'm not even sure it is a real frame so let's just remove it.
2016-07-09 10:03:21 +02:00
Gregory Hainaut 82d83ca579 gsdx: dull driver
Hopefully futur low-end GPU will get 4GB of VRAM
2016-07-08 21:53:42 +02:00
Gregory Hainaut 3f03f7333c gsdx: improve builtin profiler
* Does the first vsync (start counter) after the sleep
* Dump data after the rendering, avoid to count extra destructor,sleep time
* Dump data into a basic csv file (if people want nice graph)
2016-07-08 21:47:53 +02:00
Gregory Hainaut 8b3e04d1b6 onepad: fix keyboard on 2nd pad
There is only a single event queue, so you need to detect the pad based
on the configuration

Mouse/Wiimote is limited to first pad

Related to issue #1441
2016-07-08 19:07:42 +02:00
Gregory Hainaut eefe3e8d4f gsdx ogl: reserve 200MB of the VRAM for various gl buffers (IBO/VBO/PBO/UBO) 2016-07-08 09:43:47 +02:00
Gregory Hainaut df4645a84a gsdx ogl: Don't use NV extension on AMD
potentially the NV extension is wrongly used but can't test it with the free driver
2016-07-07 23:48:56 +02:00
Gregory Hainaut d6e447a89c gsdx: fix for old gl header release (build bot) 2016-07-07 22:23:20 +02:00
Gregory Hainaut d8050634f6 gsdx ogl: plug GL_NVX_gpu_memory_info and GL_ATI_meminfo
to query available memory on the GPU

Requirement
* dedicated GPU (ofc!)
* proprietary driver. Free driver supports r600, radeonsi
2016-07-07 22:05:15 +02:00
Gregory Hainaut 7e2b3da928 gsdx ogl: monitor global GPU memory texture usage
So far there is a hardcoded limit at ~3.7GB

In the future, the limit will be reduced for low-end GPU.
2016-07-07 22:05:15 +02:00
Gregory Hainaut f450056626 gsdx: fix initialization list order
In file included from GSRenderer.cpp:23:0:
GSRenderer.h: In constructor ‘GSRenderer::GSRenderer()’:
GSRenderer.h:58:12: warning: ‘GSRenderer::m_dev’ will be initialized after [-Wreorder]
  GSDevice* m_dev;
            ^
GSRenderer.h:52:13: warning:   ‘GSVector2i GSRenderer::m_real_size’ [-Wreorder]
  GSVector2i m_real_size;
             ^
GSRenderer.cpp:32:1: warning:   when initialized here [-Wreorder]
2016-07-07 22:05:15 +02:00
Gregory Hainaut a735e2b58f Merge pull request #1287 from mogaika/debug_window_improve
Debugger features
2016-07-07 19:59:33 +02:00
Gregory Hainaut 29c97a9bf2 gsdx ogl: only enable accumulation hack in HDR algo
Goal is to reduce shader permutations and improve perf when sw blending is disabled
2016-07-07 19:56:23 +02:00
Gregory Hainaut 16c2baa0df gsdx-ogl: don't hardcode the accumulation blend trick
Perf impact is bigger than expected
2016-07-07 19:56:23 +02:00
ramapcsx2 6b6821d0c6 Merge pull request #1433 from FlatOutPS2/master
Merge Circuit: Enhance offset detection of output circuit
2016-07-06 12:15:08 +02:00
Catarax 31a74abebe GameDB: Kessen 2 refraction patch (NTSC-J/NTSC-K) (#1447)
Apply the refraction patch for Kessen 2 NTSC-J/NTSC-K releases.
2016-07-05 09:13:17 +01:00
ramapcsx2 5c614bceba Merge pull request #1446 from Catarax/master
GameDB: Syphon Filter (The Omega Strain) NTSC-K fixes
2016-07-04 14:45:20 +02:00
Akash 6085da91e2 GSDX-Dialog: Clamp custom res and sw threads values to input range (#1443)
Fixes crash on custom resolution when users select a value below 256 or above 8192
2016-07-03 22:04:16 +01:00
Vladimir Jigulin 7e91eb6616 debugger: some user-friendly improves in ui
Grouping bytes in debugger memory window, following pointers and history
Goto in register view
Printing strings in pointer registers
Memory view can be resized correctly by ctrl+wheel
Improvement in function 5900DebugInterface::isValidAddress
2016-07-03 22:01:09 +03:00
Catarax 8aa70f05ad GameIni: Syphon Filter (The Omega Strain) NTSC-K fixes
NTSC-U fixes of Syphon Filter (the Omega Strain) applied to the NTSC-K release.
2016-07-03 20:14:18 +02:00
Gregory Hainaut a1a5f470f7 gsdx-ogl: fix builtin profiler frame counting 2016-07-03 15:54:58 +02:00
Gregory Hainaut cc62e8c785 Merge pull request #1439 from ssakash/Cleanup_Warnings
PCSX2: Clean up warnings on MSVC
2016-07-03 15:45:39 +02:00
Akash d69f2b6cc5 cdvdGigaherz: Clean up warnings on MSVC
SectorConverters: convert variables to u8.
2016-07-02 23:38:02 +05:30
Akash 1bcb5e0cc1 dev9ghzdrk: Clean up warnings on MSVC
pcap_io: cast ``pcap_io_running`` to bool.
2016-07-02 23:38:01 +05:30
Akash 264eb52d61 USBqemu : Clean up warnings on MSVC
usb-kbd: Remove unused variable.
usb-ohci: Add proper casts for the variables.
vl: Add proper casts for the variables.
USB: Add proper casts for the variables.
2016-07-02 23:37:37 +05:30
Gregory Hainaut 11eeeb6ab1 gsdx ogl: be sure sw blending is enabled in sw colclip
Hit the assertion on superman
2016-07-02 17:19:41 +02:00
Akash 4cfb2b248e GSDX: Clean up warnings on MSVC
GLLoader: cast passed parameters to required type.
GSDeviceOGL: cast variables to required type and silence warnings.
GSRendererOGL: cast variables to required type and silence warnings.
2016-06-30 16:50:10 +05:30
Akash 72661e7c16 PCSX2-Core: Clean up some warnings on MSVC
x86emitter : Convert variable type from u8 to bool.
recVTLB: Cast "sign" to bool to prevent a warning.
R5900OpcodeImpl: Cast all the values in array to u64 instead of s64.
2016-06-30 16:49:18 +05:30
Gregory Hainaut a065a1d918 gsdx ogl: don't use GL_ARB_clear_texture for the depth buffer
It creates some slowdowns for unknown reason. My best hypothesis is
that stencil will be cleared too which is slow.

Let's keep the code for the future when stencil will be dropped.

Fix #1420
2016-06-30 09:40:07 +02:00
FlatOutPS2 1e3e343e43 Merge Circuit: Enhance offset detection of output circuit
If baseline and display rectangle offsets differ by small values then consider the status of frame memory offsets, prevents blurring on Tenchu: Fatal Shadows, Worms 3D, ProStroke Golf, Vexx
2016-06-29 22:35:53 +02:00
Jonathan Li 79d019b5bb utilities: Don't use TLS buffers in FastFormatString classes (#1430)
The TLS buffers used by the FastFormatUnicode and FastFormatAscii
classes seem to be responsible for PCSX2 not terminating properly on
Windows under certain conditions (using MTVU before commit
1111e03901, using CDVDgigaherz without a
disc, possibly other conditions).

When PCSX2 shut downs and the FastFormatBuffers are being cleaned up,
the call to pthread_key_delete() would end up calling
WaitForSingleObject(e, INFINITE) and waiting indefinitely for an event
to trigger. It never does get triggered (for reasons unknown) and
therefore PCSX2 doesn't terminate properly.

Remove the usage of TLS buffers in the FastFormatString classes - it
fixes the termination issue on Windows and doesn't seem to have much
effect on performance.
2016-06-29 13:13:51 +01:00
Gregory Hainaut 1f4f55bcc9 gsdx: tentative fix for 64 bits buildl 2016-06-26 18:27:50 +02:00
Gregory Hainaut 4ec941440d gsdx replay: use the new profiler data 2016-06-26 17:53:22 +02:00